Research On Predicting Quality Of Software Services In An Open-source Environment

The development of service-oriented software architecture has promoted the openness of software development patterns.Different from the previous development pattern,it's more convenient for developers to use Web services online or to publish their own services.But the number of services with the same or similar function is growing rapidly in this open developing environment.Users not only demand for specific functions,but also emphasize high quality of software services.But the big amount of similar services makes it difficult to find the ones with high quality that are best suitable for target users.Also it's a great challenge for services providers to improve the quality to attract more users.It will benefit a lot if we can predict the quality of services before invoking.The quality of service is about the non-functional features of the software service including internal quality and external quality.The internal quality refers to the inherent attributes of the service itself,which are not affected by external interactive environment.The defects are commonly used to measure the quality of internal quality.The external quality refers to the service performance attributes observed by users when invoking the service,such as response time,throughput rate.Unlike internal quality,the external one is also relevant to the users.The prediction on internal and external quality will have a great significance to users,providers,and service platforms.But there are still some shortcomings in the research on quality prediction of software services,including the following three aspects.(1)The researchers are mainly focus on the defect-proneness of Web service rather than the defect numbers.And there are some relevant issues which are not clear.(2)For the prediction of QoS,the rating oriented approaches are mainly trying to predict the QoS attribute value in user-item matrix as accurate as possible.But the more accurate rating predictions are not always lead to better recommendation.The ranking methods are more suitable for the service recommendation.(3)The guidance of design and implementation of QoS aware service recommendation system remains so scarce.In order to solve the problems mentioned above,we conduct researches on defect numbers prediction,QoS prediction and QoS aware service recommendation.The main work and contributions are shown as follows.(1)The defect-proneness predictions in software modules offer a great understanding of internal quality of services to users,but also enable the developers to deal with the modules with higher defect risk first,which make the software testing process more efficient.However,the useful information is limited from defect-proneness prediction.With defect numbers prediction,we can achieve even greater gains.In this paper,we investigate the feasibility of defect numbers prediction.And the problems within it,such as which regression method is best for defect numbers prediction.The empirical experiments on PROMISE repository show that the cross projects predictions can be as good as the within projects ones with proper training dataset,and the random forest regression model is the best for defect numbers prediction.(2)We propose a rating and ranking based hybrid recommendation model(2RHyRec)to predict the services ranking order.The model combines the benefits of neighbor based collaborative filtering and latent factor models at analyzing the user service matrix of QoS value.And we take the ranking based metric to evaluate the models.The experimental results on real Web service data repository WSDream show indicate that the hybrid recommendation model proposed in this paper outperform the others.(3)Based on the hybrid recommendation model,we build a QoS-aware personalized Web service recommendation system.In this system,after the user specifies the functional requirements,it will recommend the services with best QoS based on the historical invoking records.More details about design and architecture are shown in the paper.In general,we provide a feasible solution for the realization of the personalized recommendation system.